MasterOdds

Why it is paused

Most prediction products answer 'who will win?', which is the easy half and the useless half. MasterOdds was built to answer the harder one: is the price the market is quoting actually wrong, and by how much?

It builds a rating backbone for every international team by format and gender, adds point-in-time form, venue and toss features, corrects with a gradient-boosted model and then calibrates the output — so that when it says seventy percent, it means seventy percent. The fair-probability model is never shown market odds, on purpose: the moment it sees the price it starts predicting the price instead of the match. Only after that does it strip the bookmaker's margin out of the quoted odds and compare the two.

It was backtested the honest way — walk-forward, out of sample, across 9,300+ international matches from 2001 onward. On limited-overs cricket it called 68% of them correctly, with a near-diagonal reliability curve: when it says seventy percent, it wins about seventy percent of the time. Calibration, not accuracy, is the number that matters for a product like this, and that is the one we tuned for.

Watching it against live matches himself, Aby's own tally ran higher — around 84%. That is a founder's record over a short run, not an audited backtest, so the 68% is the figure this page stands behind.

It is paused. Not for the legal reason people assume when they see cricket and odds in the same sentence — it takes no bets and runs no market — but because being good at this raises a question about what kind of company we want to be, and that is a decision, not a bug.